ABM Validation Report
Pharma Supply Chain Agent-Based Model — Confidence Assessment300 Monte Carlo runs • 30 seeds • 5 scenarios • Generated 4/21/2026
| Scenario | Config | Avg SL | Min SL (P50) | Min SL P10 | Min SL P90 | Std Dev | Avg Stockouts | Terminal SL |
|---|---|---|---|---|---|---|---|---|
| India API Export Ban | AI-Enabled | 80.6% | 55.7% | 52.0% | 58.7% | ±2.6% | 179 | 89.1% |
No-AI | 79.4% | 53.0% | 48.6% | 56.3% | ±3.0% | 182 | 87.3% | |
| China Raw Material Lockdown | AI-Enabled | 80.8% | 50.8% | 46.8% | 53.8% | ±2.6% | 149 | 89.3% |
No-AI | 79.1% | 48.6% | 42.7% | 53.5% | ±3.9% | 152 | 86.9% | |
| US Hurricane | AI-Enabled | 88.3% | 58.1% | 36.5% | 78.2% | ±14.9% | 170 | 89.4% |
No-AI | 86.6% | 53.1% | 30.6% | 74.4% | ±13.5% | 174 | 87.1% | |
| Cyber Attack | AI-Enabled | 93.7% | 88.5% | 82.8% | 92.3% | ±4.5% | 146 | 89.4% |
No-AI | 92.5% | 81.8% | 73.2% | 85.9% | ±6.7% | 157 | 88.0% | |
| Quality Crisis | AI-Enabled | 77.1% | 59.5% | 56.2% | 62.5% | ±2.7% | 137 | 68.3% |
No-AI | 75.6% | 53.2% | 47.5% | 56.7% | ±3.1% | 141 | 65.6% |
Service Level (%) — AI vs No-AI
Total Inventory — AI vs No-AI
Agent-Based Model
Multi-tier pharmaceutical supply chain with Tier 3 (raw materials) → Tier 2 (API synthesis) → Tier 1 (formulation) → Manufacturing → Distribution → Pharmacy agents. Each agent has stochastic production, inventory, and shipping behaviors governed by a seeded LCG PRNG for reproducibility.
Monte Carlo Design
30 independent seeds × 5 disruption scenarios × 2 configurations (AI-Enabled, No-AI) = 300 total simulation runs over a 52-week horizon. Disruptions begin at week 2.
Disruption Types
Trade Dispute: Cascading export ban with inventory seizure. Pandemic Wave: Sustained multi-tier lockdown with bi-weekly drain. Natural Disaster: Catastrophic physical destruction with fast V-shaped recovery. Cyber Attack: Brief digital disruption, no inventory destruction. Quality Failure: Rolling recall waves every 5 weeks.
Validation Criteria
8-point scorecard assessing scenario differentiation, AI vs No-AI separation, recovery dynamics, Monte Carlo convergence, and absence of pathological model behaviors. Generated 4/21/2026.